Scholarlink

<p>This work presents a practical, cost-effective pipeline to predict the hardness of hot-forged bainitic steels directly from process parameters, bringing AI techniques closer to industrial deployment. We combine wedge-forging experiments with inverse FEM to generate a database from which thermomechanical descriptors (equivalent strain, strain rate, and initial and final forging temperatures) and the corresponding hardness values were extracted. We benchmarked Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Random Forests (RF). The ANN, configured with three hidden layers (64, 128, and 256 neurons), achieved the best performance, with a Mean Absolute Error (MAE) of 1.3518 HV, a Mean Squared Error (MSE) of 8.8448 HV², and a coefficient of determination (R²) of 0.994. Inference tests on forged connecting rods demonstrated robustness and transfer to real components, underscoring the model’s potential as a soft sensor for decision support and supervisory process adjustments. By coupling minimal, physically meaningful inputs with high predictive accuracy, the study lowers barriers to AI adoption in hot forming and offers a tool with potential for early drift detection, parameter optimization, and improved quality and operational efficiency in industrial forging.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predictive modeling of hardness in bainitic steel forging processes using artificial intelligence

  • 期刊论文

  • 发表日期：

    2025年11月3日

查看全文
  • Peterson Duarte Diehl,
  • André Rosiak,
  • Roderval Marcelino,
  • Lirio Schaeffer

摘要

This work presents a practical, cost-effective pipeline to predict the hardness of hot-forged bainitic steels directly from process parameters, bringing AI techniques closer to industrial deployment. We combine wedge-forging experiments with inverse FEM to generate a database from which thermomechanical descriptors (equivalent strain, strain rate, and initial and final forging temperatures) and the corresponding hardness values were extracted. We benchmarked Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Random Forests (RF). The ANN, configured with three hidden layers (64, 128, and 256 neurons), achieved the best performance, with a Mean Absolute Error (MAE) of 1.3518 HV, a Mean Squared Error (MSE) of 8.8448 HV², and a coefficient of determination (R²) of 0.994. Inference tests on forged connecting rods demonstrated robustness and transfer to real components, underscoring the model’s potential as a soft sensor for decision support and supervisory process adjustments. By coupling minimal, physically meaningful inputs with high predictive accuracy, the study lowers barriers to AI adoption in hot forming and offers a tool with potential for early drift detection, parameter optimization, and improved quality and operational efficiency in industrial forging.